Simon Willison is the guy behind a surprising amount of the AI vocabulary you’ve read here. He named Prompt Injection in 2022, the Lethal Trifecta in 2025, and he helped make “slop” the word for junk AI content. He doesn’t run an AI lab. Instead, Willison writes a blog, tests everything himself, and says out loud what the companies selling AI would rather not.
If an AI security story makes sense to you, there’s a decent chance he explained it first.
Who they are
Willison is a British programmer who lives in Half Moon Bay, California. Back in 2003, while working at a small newspaper in Lawrence, Kansas, he co-created Django, a tool for building websites that’s still widely used today. Later he co-founded a startup that Eventbrite bought, and then spent a year at Stanford as a journalism fellow.
These days he builds free tools (Datasette for exploring data, and a command-line tool called LLM for talking to AI models) and posts on simonwillison.net, which he’s run since 2002. Readers fund him directly through GitHub Sponsors, which helps him stay independent of the companies he writes about.
Why they matter
He gives problems names. In September 2022, after another engineer posted about tricking GPT-3 with hidden instructions, Willison wrote it up and called it prompt injection, borrowing from an old hacking term. Once a problem has a name, people can actually talk about fixing it.
He’s blunt about the fix. He’s argued for years that nobody has a reliable way to stop prompt injection. His advice is to limit what an AI agent can reach, since nobody can fully trust what it reads.
People who build AI listen to him. He’s not selling a model or a security product, so builders treat his testing as a fairly honest read on what new tools can and can’t do. Your Data Exfiltration and Jailbreaking vs. Prompt Injection cards both lean on his work.
What they’ve said or done
In June 2025, he described the Lethal Trifecta: an AI agent with access to your private data, exposure to untrusted content, and the ability to send information out. Not understanding that combination, he wrote, “can let an attacker steal your data.”
Then he added the line that stuck: “The LLM vendors are not going to save us!” Coming from someone who uses these tools every single day, that one’s worth sitting with.

